Abstract: In view of the importance of self-protected flux-cored wire welding in the field, especially for the emergency repair of large machinery and other key parts in the field, it is very important to improve its welding forming accuracy. Therefore, the process stability of self-protected flux-cored wire in double-pulse arc mode is studied. In order to control the stability of arc welding process effectively, the influence of double pulse parameters on the stability of arc welding process was studied by single factor experiment. It was found that the average droplet size was closely related to the current variation coefficient. Based on this, 7 double pulse parameters, including strong and weak pulse frequency, duty cycle, peak current and base current of strong and weak pulses, were selected as input values, and a prediction model of the average droplet size based on RBF-BP neural network was established. The results show that the prediction model is effective and feasible, and provides technical support for controlling the stability of the droplet transition process.